JP3393494B2 - User dictionary creation support device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language translation system for automatically translating a sentence in a first natural language into a second natural language, and more particularly, to a user. The present invention relates to a device that supports the creation of a user dictionary that gives a translation of a second natural language to a word or expression of a first natural language that is used individually. 2. Description of the Related Art In general, the creation of a user dictionary used in a natural language translation system is to be performed at the responsibility of the user, and what to register (a phrase in the first natural language), Basically, the system does not concern whether to translate (translation phrase of the second natural language) or what the expression is (word attribute). However, some commercially available natural language translation systems can provide reference information for creating a user dictionary. For example, when the natural language translation system is run in a specific mode, a list of words (unknown words) unknown to the system is output, or an online dictionary search is performed to determine an expression to be translated. [0003] In the above-mentioned prior art, it is possible to provide support for selecting a phrase in the first natural language and determining a translated phrase in the second natural language. However, it is necessary for the user to individually determine the word attribute based on a classification table previously published by the developer of the natural language translation system. In terms of the accuracy of a natural language translation system,
It is known that if the classification table is separated to a certain degree, precise interpretation can be performed ("Ikehara, Miyazaki, Yokoo; Dictionary of Semantic Interpretation for Japanese-to-English Machine Translation, Information Processing Laboratories NL-84").
-13, July 1991 points out the necessity of about 3000 word classifications), and it is extremely burdensome for a general user to determine word attributes according to such fine classification. Also, for developers of natural language systems, from the viewpoint of security related to the know-how of classification, it is sometimes desirable to only publish the classification table itself to a level of classification that is coarser than a certain level. Attributes cannot be set at a fine level. [0005] For this reason, regarding information registered in the system dictionary, when it is desired to change only the first natural language phrase or the second natural language translation, or register a long phrase that includes the registered word as a partial expression. If the user wants to do so, the user sets the word attribute, and depending on the setting, the structure of the translated sentence changes drastically, which causes the translation quality of the natural language translation system to deteriorate. SUMMARY OF THE INVENTION It is an object of the present invention to reduce the burden on a user required to create a user dictionary used for a natural language translation system and to publish a classification table required when setting word attributes of the user dictionary. It is another object of the present invention to provide a user dictionary creation support device that can keep the user dictionary at a coarse level. [0007] A user dictionary creation support apparatus according to the present invention comprises: a system dictionary holding a first natural language phrase, a second natural language translation, and a word attribute; When the user has entered in the system dictionary a phrase with the same notation as the entire first natural language phrase entered by the user for the pair of the first natural language phrase and the second natural language translated phrase Extracts the word attribute for the phrase from the system dictionary, and if a translation having the same notation as the whole translation of the second natural language input by the user is registered in the system dictionary, the phrase giving the translation Is extracted from the system dictionary, and the partial expression including the central word of the first natural language phrase input by the user matches the entire registered phrase of the system dictionary or the partial expression including the central word of the registered phrase. You In this case, the word attribute corresponding to the matched phrase is extracted from the system dictionary, and the partial expression including the central word of the translation of the second natural language input by the user is the whole or the registered translation in the system dictionary. If the subexpression including the central word of the translated phrase matches, the word attribute of the expression giving the matched translated phrase is extracted from the system dictionary, and the word attribute of the first natural language phrase input by the user is extracted. It has an attribute estimating unit that automatically estimates the result obtained by the estimation and reflects the result in the user dictionary or presents the result to the user, and then reflects the result in the user dictionary based on the selection result of the user. According to the present invention, a pair of a first natural language phrase and a second natural language translation phrase input by a user is compared with the first natural language phrase stored in the system dictionary. Alternatively, in the case where there is a phrase that entirely or partially matches the translated phrase of the second natural language, the corresponding word attribute is extracted from the system dictionary, so that the burden on the user is reduced. The word attributes automatically extracted as described above do not necessarily need to be presented to the user to a fine level. If the classification is performed hierarchically, the word attributes are internally set finely. However, it is possible to present to the user at a coarse level, which is effective from the viewpoint of security. Next, an embodiment of the present invention will be described in the case where the first natural language is Japanese and the second natural language is English. FIG. 1 is a block diagram of a user dictionary creation support device according to an embodiment of the present invention. Attribute estimator 2 for Japanese words and English translations input from terminal 1
Refers to the Japanese-English bilingual dictionary 3, which is a system dictionary used for the Japanese-English translation system, estimates word attributes, displays them on the terminal 1 as required, requests correction of the user, and receives the operation of the user. The input bilingual information and word attributes of Japanese and English are registered in the user dictionary 4. It is composed of an attribute estimator 22, a Japanese partial match attribute estimator 23, and an English partial match attribute estimator 24. FIG. 2 is a flowchart showing the operation of the user dictionary creation support device of FIG. As shown in Table 1, the English-Japanese bilingual dictionary 3
The operation of the user dictionary creation support device of FIG. 1 will be described for the case where "t" is registered together with the word attribute "<< treatment >>". [Table 1] When the user inputs “treatment → cure” (Step 3
1) In the Japanese-English bilingual dictionary 3, the Japanese part is the same word "treatment"
Therefore, the word attribute << treatment >> is taken out (steps 32 and 36), presented to the user to receive an instruction (step 37), and registered in the user dictionary 4 as shown in Table 2 (step 37). Step 38). [Table 2] Next, when the user inputs “treatment → treatment” (step 31), the Japanese-English bilingual dictionary 3 does not have the same word in the Japanese part, but has the same word in the English part. (Steps 32, 33, and 36), and are registered in the user dictionary 4 in the same manner as above (Steps 37 and 3).
8). When the user inputs "radiation therapy → radiotherapy" (step 31), the Japanese-English bilingual dictionary 3 has no word that matches both the Japanese part and the English part. Is found, the word attribute is extracted (steps 32, 33, 34, 36) and registered in the user dictionary 4 in the same manner (steps 37, 3).
8). When the user inputs "medical treatment-> medical treatment" (step 31), the Japanese-English bilingual dictionary 3 has no word that matches both the Japanese part and the English part, but the English central word "treatment" is not. Since there is a matching word, the word attribute is extracted (steps 32 to 36) and registered in the user dictionary 4 in the same manner (steps 37 and 38). When "Numerically controlled robot → numerical controlled robot, << industrial machinery >>" is registered in the system dictionary 3, if the user inputs "computer controlled robot → computer controlled robot", the Japanese part will be displayed. Since the expression "controlled robot" or the English partial expression "controlled robot" matches, the word attribute can be extracted also in this case. In particular, when a plurality of corresponding words are extracted, the result is presented to the user and made to be selected by the user, thereby giving highly accurate word information while keeping the user's burden small. Can be. The word attributes extracted from the Japanese-English bilingual dictionary 3 may be registered in the user dictionary 4 without being presented to the user. Although not mentioned here, as a method of extracting the central word, a Japanese compound word is modified from the front only with respect to the central word "robot" like "numerically controlled robot", and an English compound word is obtained. Then, "numerical controll
The central word "robot" is modified from the front like "ed robot" and the central word "desk" is modified from the back by phrase or clause like "desk with lighting unit". In any case, the center word can be estimated relatively easily. As described above, according to the present invention, the system holds the first natural language phrase and the second natural language translation pair entered by the user. If there is a whole or a partial match with a natural language phrase or a second natural language translation, the user can use the natural language translation system by extracting the corresponding word attribute from the system dictionary. In addition to reducing the burden on the user required to create the dictionary, there is an effect that the classification table necessary for setting the word attribute of the user dictionary can be kept at a coarse level.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a user dictionary creation support device showing one embodiment of the present invention.
It is a block diagram of a device. FIG. 2 is a flowchart showing an operation of the user dictionary creation support device of FIG. 1; [Description of Code] 1 Terminal 2 Attribute Estimator 3 Japanese-English Bilingual Dictionary 4 User Dictionary 21 Japanese Match Attribute Estimator 22 English Match Attribute Estimator 23 Japanese Part Match Attribute Estimator 24 English Part Match Attribute Estimator 31-38 steps

──────────────────────────────────────────────────続 き Continuation of front page (72) Inventor Satoru Ikehara 1-6, Uchisaiwaicho, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation (56) References JP-A-4-177463 (JP, A) (58) Field surveyed (Int.Cl. ⁷ , DB name) G06F 17/21-17/28 JICST file (JOIS)

Claims (1)

(57) [Claims] [Claim 1] In a natural language translation system for automatically translating an input sentence of a first natural language into a sentence of a second natural language, a first language used by a user individually A user dictionary that provides a translation of a second natural language to a word or expression of one natural language, and supports the creation of a user dictionary that holds words and phrases of the first natural language, translations and phrases of the second natural language, and word attributes A creation support device , comprising: a system dictionary holding a first natural language phrase, a second natural language translation, a word attribute; a first natural language phrase and a second When a phrase having the same notation as the entire phrase of the first natural language input by the user is registered in the system dictionary for a pair of translated words of the two natural languages, the word attribute of the phrase is changed to the system dictionary. From the second natural language input by the user When a translation having the same notation as that of the entire translation is registered in the system dictionary, a word attribute of the phrase giving the translation is extracted from the system dictionary, and the word attribute of the first natural language phrase input by the user is extracted. If the partial expression including the central word matches the entire registered word in the system dictionary or the partial expression including the central word of the registered word, the word attribute for the matched word is extracted from the system dictionary. If the partial expression including the central word of the translation of the second natural language input by the user matches the entire registered translation of the system dictionary or the partial expression including the central word of the registered translation, the matching translation By extracting a word attribute for the expression giving the phrase from the system dictionary, the word attribute for the first natural language phrase input by the user is estimated. The result obtained, either automatically reflected in the user dictionary, after presenting to the user, the user dictionary creation support apparatus having an attribute estimation unit configured to reflect the user dictionary based on the user's selection results.